How to find blog topics from what ChatGPT and other AI assistants say about your category

You find blog topics from AI answers by asking the assistants the questions your buyers ask, then looking at three things in each answer: whether you're named, who is named instead, and which pages the answer was built from. ChatGPT, Gemini, Perplexity, and Claude often show the pages they drew on when they search the web, and those cited pages are the most useful part of the answer. They show you what kind of page the assistant trusts for that question, and whether it's a page you could write.

The quickest way to see where you stand today is Kite's free AI visibility checker. It asks ChatGPT, Gemini, Perplexity, and Claude the questions your customers ask, and in about 45 seconds it shows whether they recommend you and who they recommend instead. It's free and needs no account, so it's a good first look before you plan any posts.

To do the full job by hand, start with 25 or more questions a buyer might ask before they've heard of you, such as "What's the best way to handle X for a small team?" or "What should I look for in a Y tool?" Leave your own brand out of them, because the point is to see whether the assistants bring you up on their own. Ask each question in a fresh chat in each assistant, and ask it more than once, because the answers change from one run to the next. Then log every answer in a spreadsheet with these columns:

• The question and the assistant you asked
• Whether you were named, and how you were described
• The other brands named, in order
• Every page the answer cited, with its address
• Anything the answer got wrong, left out, or hedged on

That spreadsheet tells you where you're missing. It doesn't tell you what to write yet. For that, you have to open the cited pages and see what they contain. If the answer to "best Y tools for startups" cites three roundup posts and none of them includes you, the topic might be your own roundup, or it might be a pitch to get onto those three lists. If the answer cites a competitor's comparison page, the topic is your version of that comparison, done more fairly and more usefully. If the answer describes your category in a way that's out of date, the topic is a clear explainer that says how it works now.

A good result is a short list of topics, each tied to the specific questions it's meant to win, the page it's meant to replace or beat, the format that question rewards, and a way to check later whether it worked. AI visibility tracking tools can automate the asking and logging across many questions. Reading the cited pages, deciding which gaps a blog post can close, and briefing the posts is still the work that turns the data into a writing plan.

Kite does that whole job, from choosing the questions to measuring whether the published post changed the answers.

Start by asking Kite for topics in plain words:

"Find blog topics from what ChatGPT and the other AI assistants say when people ask about our category."

Kite is an AI marketer. It researches your business, builds a growth strategy, and ships the work that brings you new customers. By the time you ask, it already knows your positioning, your customers, and your competitors from that research, so it can propose the subjects to measure and the buyer questions to ask without a long brief. You can work with Kite in your browser or in Slack.

The first time Kite shows you its findings, it states how it understands your positioning and asks whether that's right. Your answer guides every later judgment, such as whether an answer describes you correctly. It doesn't ask again unless you change your positioning.

To track your standing over time, your website needs to be in your Kite workspace with its domain set, whether Kite built it or you imported it. If you only want a one-time look at the current standings, Kite can read them for any domain. You don't need an account with any AI visibility tool, because Kite runs the questions through ChatGPT, Gemini, Perplexity, and Claude itself.

Kite chooses questions that match how your buyers decide, not just the words they type into Google. It starts from what it knows about your company: what you offer, who you serve, which buying decisions matter most to you, and which demand you can realistically win. From there it writes at least 25 distinct questions, and up to about 80 when your market supports them, across the different problems and decisions a buyer goes through. It avoids near-duplicate wordings, keeps your brand name out of them, and names a competitor only when the question is a real comparison a buyer would make.

You can add your own questions, and Kite asks them word for word every time it measures. If you'd like suggestions, ask Kite to find questions worth tracking, and it lists candidates in Kite for you to tick. It also keeps a list of subjects, such as "website audit tool" or "how to improve SEO," that it generates more questions from.

Kite asks every question in ChatGPT, Gemini, Perplexity, and Claude and saves each answer with its citations. For each question it records whether your brand appears, which competitors appear, and which pages and domains the answer cites. Across the whole set, it ranks your brand against every brand the assistants mention and measures your share of voice, which is how often you're named compared with the others.

It also brings in the search side. For the keywords your site ranks for or wants to rank for, Kite records the monthly search volume, your current position, and whether Google shows an AI Overview. That lets it spot the highest-value gaps: topics people search for often, where an AI answer already exists, and where you don't appear in it.

Kite reads the pages the assistants cite most, because the brand list alone doesn't explain why a competitor wins. For roughly the six most-cited pages, it opens each one and records which of the brands in the answers it names, whether it names you, and what kind of page it is: a roundup, a comparison, product documentation, or a community thread. A page Kite can't open is marked unverified rather than guessed at.

This is the step that turns a scoreboard into topics. When two or three competitors above you all appear on the same kind of page, Kite names that shared pattern, because it's the strongest clue to what the assistants trust for those questions.

Kite sorts what it found by the kind of gap, because each kind calls for a different response. It looks at six things:

• How often you appear compared with your competitors
• Whether you're described the way you want to be, or as something else
• Whether you show up for the subjects you want to own
• Which kinds of pages the assistants cite for those subjects, and whether you have that kind of page
• Whether you're on the outside pages the assistants keep citing, such as roundups and review sites
• Whether answers name specific brands at all, or only describe the category in general

Some of these gaps are blog topics and some aren't. Missing the page type that gets cited, such as a comparison or a buyer's guide, is a topic. So is being described wrongly or vaguely, which a clear explainer on your own site can correct. Missing from third-party roundups is usually not a topic for your blog, so Kite recommends a pitch to those publications instead and drafts it for you to send. It also notes where you already appear, so you know which answers to protect while you work on the rest.

Kite turns the gaps into ten ranked recommendations, plus a few alternates held in reserve, and each one it recommends writing comes as a brief complete enough to write from. It ranks them mainly by how much each could change your visibility, and uses how easy each is to do to break ties. A post aimed at a keyword with real search volume and an AI Overview you're missing from ranks above one with no measurable demand.

Each topic brief contains:

• The working headline, with the subject named up front so an assistant can quote it
• The claims the post has to make, written so the first paragraph answers the buyer's question directly
• The elements the page needs, such as a comparison table naming the competitors from the answers, a short questions section, or structured data
• The exact questions the post is meant to win, quoted word for word
• The evidence behind it, such as which competitors appear on which cited pages
• The result to watch for, such as appearing in the answers to two named questions within the next two measurements
• Any honest limit, such as a post that should wait until a site problem blocking that page is fixed

Each recommendation also says why its approach works, not just that competitors do it. Kite notes which assistants a recommendation targets when the reason is specific to one of them, since each leans on different kinds of sources.

When you tell Kite to go ahead with a topic, it writes the post from that brief in your brand's voice. On a site Kite hosts, or one connected through a supported content system such as Webflow, Contentful, Sanity, Storyblok, or self-hosted Payload, Kite prepares the post for you to review and publishes it once you approve. For other sites, you get the finished post to publish yourself. Kite records your decision on every recommendation, including the ones you turn down, so a topic you dismissed doesn't come back under a new name.

Kite checks whether the post worked by asking the same questions again. For each recommendation it carried out, it compares the four weeks after the work finished with the four weeks before, counting how many of the answers to the post's target questions named you. For the page itself, once Google's figures are final, it compares clicks, impressions, click-through rate, and average position for the four weeks on either side of publication. Small movements are reported as within normal variation rather than as a win or a loss.

New measurements feed the next round of topics. When the answers keep coming back to a subject your list doesn't track yet, Kite adds it so future measurements cover it.

When Kite finishes, you have a writing plan built from what the assistants actually say, the posts you approved, and a way to see whether each one moved the answers.